Research On Proactive Scheduling Approaches For Job-shops Based On Sensory Data In Wisdom Manufacturing

Under the background of Cloud Computing,Internet of Things(Io T),Big Data,Cyber-Physical Systems,Enterprise 2.0 and Industrie 4.0,along with the deep fusion of information technology and advanced manufacturing technology.Wisdom Manufacturing(WM)has been put forward based on Socio-Cyber-Physical Systems(SCPS),which forms a service-oriented manufacturing model of humans-computers-things collaboration based on knowledge.In WM environment,the entire production workshop is covered by Io T.Various sensors such as RFID and accelerometers are deployed in the production workshop,the whole production process is monitored in real time,and the sensed data is sent to the processing center.Abnormal events often occur due to various uncertain factors during the production process,which results in complex production information and hard to control.It is needed to process varieties of sensory data in real time,to dig abnormal events out of production fields and to predict abnormal events which will occur,Thus,proactive scheduling is implemented based on the real-time and prognostic abnormal events,so that the damage to the production system is avoided.To this end,this thesis mainly focuses on the proactive scheduling based on the abnormal event monitoring of workpieces and remaining useful life prognostic of tools,including the following contents:(1)Wisdom manufacturing modelThe characteristics of wisdom equipment are analyzed and summarized.WM is addressed under the perspectives of network convergence and SCPS,and the social environment and key technologies of WM are discussed.(2)Abnormal event monitoring of workpieces based on RFID.The sensing-aware environment of a WM workshop is built,in which all kinds of event models for RFID are defined,including tag event,simple event and complex event.A framework of Complex Event Processing(CEP)system is given,and a synthetic RFID data cleaning method is proposed to realize abnormal event monitoring of workpieces in real time.Lastly,the data cleaning method and abnormal event monitoring are verified by the experiments.(3)Tool condition monitoring based on a wireless accelerometerA framework of Tool Condition Monitoring(TCM)system is given,and the experiment setup for TCM is built.The vibration signal is de-noised by Wavelet Transform,the features are extracted by different methods in the time,frequency and time–frequency domains,and the key features are selected based on Pearson's Correlation Coefficient(PCC).The Neuro-Fuzzy Networks(NFN)is adopted to predict the tool wear,and the Human Machine Interface(HMI)of tool wear and Remaining Useful Life(RUL)prediction is programmed.Finally,compared with the Back Propagation Neural Networks(BPNN)and Radial Basis Function Networks(RBFN),the NFN has the best performance in the prediction of tool wear and RUL.(4)Tool condition monitoring based on deep learningThe structures and training methods of five Deep Learning(DL)models are compared,and the tool wear prediction based on Convolutional Neural Networks(CNNs)is proposed.The learning platform for CNNs is built.Compared with different CNN models and the traditional Neural Networks,the proposed model has the best performance.(5)Proactive scheduling for machining job-shops in WMA classification of scheduling models is given,and the sensing-aware environment of machining job-shops is built in WM.A proactive scheduling scheme is put forward,including the mathematical model of the machining job-shop scheduling,scheduling framework,strategy and MD3 GA algorithm.The prototypical platform of a machining job-shops is built,where machines and a AGV are scheduled simultaneously,so the proposed scheduling scheme is verified by the experiments.